(update: 9/20/2006 – This is an old post I made to the blog on electorama.com, which I’m shutting down)
There was a great show about Instant Runoff on KUOW, the NPR affiliate in Seattle today. The guests were Steven Hill from the Center for Voting and Democracy as well as the Republican Party official who wrote the opposing view on the San Francisco initiative that passed. The great part was that awareness of Condorcet methods is growing. They also have a blog to post your comments.
Here’s what I posted on the site (with a couple of thinko fixes I just noticed):
Great show! A couple comments:
I think Steven Hill did a great job of explaining the value of a ranked ballot. However, in underplaying the weaknesses of Instant Runoff relative to Condorcet voting, he mislead on a couple of things:
1. It doesn’t require paradoxical voting patterns to make Instant Runoff produce an anti-democratic result. Here’s an example of Instant Runoff breaking down with a set of voters with rational preferences. In this example, Knoxville is chosen as the capital of Tennessee, even though it has weak core support and weak broad support. See the same example using Condorcet for a more rational result.
2. In close elections, Instant Runoff can result in chaos. It’s much worse than other elections. A real world example of this was when the Debian GNU/Linux project picked its project leader. They used Condorcet, and so things were fine. However, using their ballots, it was easy to calculate what the result would have been under Instant Runoff. There was a very unstable result, where removing just one vote could make any of three candidates win. Worse, the new winner (Bdale Garbee) was ranked higher on the ballot that was eliminated than the old winner (Branden Robinson). So, in essense, there’s a voter who liked Bdale Garbee better than Branden Robinson who’s vote (ranking Bdale Garbee higher than Branden Robinson) would cause Branden Robinson to win in Instant Runoff.
This is quite correct. IRV is unnecessarily complex, and has higher Bayesian regret (produces lower average voter satisfaction) than any other common voting method besides plurality. Bayesian regret calculations show that voters will be most satisfied, on average, with range voting. Here are some sample numbers from some simulations. Note that range voting is as much of an improvement over plurality as plurality is over random selection of the winner! Range voting is also much simpler than ranked order methods like IRV and Condorcet. You just score every candidate, and the candidate with the highest average score wins.
_Voting system Regret A Regret B_
Magically elect optimum winner 0 0
Range (honest voters) 0.04941 0.05368
Borda (honest voters) 0.13055 0.10079
Approval (honest voters) 0.20575 0.16549
Condorcet-LR (honest voters) 0.22247 0.14640
IRV (honest voters) 0.32314 0.23786
Plurality (honest voters) 0.48628 0.37884
Range & Approval (strategic exaggerating voters) 0.31554 0.23101
Borda (strategic exaggerating voters) 0.70219 0.48438
Condorcet-LR (strategic exaggerating voters) 0.86287 0.58958
IRV (strategic exaggerating voters) 0.91522 0.61072
Plurality (strategic voters) 0.91522 0.61072
Elect random winner 1.50218 1.00462
— http://rangevoting.org/UniqBest.html
Check out http://RangeVoting.org for more information.